beta6 upgrade

[reqshark]

@nickdesaulniers, the upgrade is done.

addresses #102 and half of #103

[thelinuxlich]

great!

[nickdesaulniers]

we don't mean to be linking in dtrace, right?

[reqshark]

nope, i'll take it out

[nickdesaulniers]

I get an error with one test:

git submodule sync
git submodule update
npm i
npm t
...
/Users/Nicholas/code/node/nanomsg/node_modules/tape/index.js:75
        throw err
                  ^
Error: stream.push() after EOF
    at readableAddChunk (_stream_readable.js:149:15)
    at Socket.Readable.push (_stream_readable.js:126:10)
    at Socket.<anonymous> (/Users/Nicholas/code/node/nanomsg/lib/index.js:207:24)

[reqshark]

are you sure you're pulling from the right submodule? what's the last commit hash on that submodule?

[nickdesaulniers]

* 5515f96 - (HEAD, tag: 0.6-beta) Bump ABI version for 0.6-beta. (4 weeks ago) <Garrett D'Amore>

[reqshark]

you're using node 0.12 aren't you

[reqshark]

using node 0.12 i'm able to reproduce your result, apparently returning after the survey test passes (and just before it completes).

✔  bent@osx /Users/bent.cardan/reqshark/node-nanomsg  [beta6|✔ ] 
12:53 $ node test/survey
TAP version 13
# inproc socket survey
ok 1 (unnamed assert)
ok 2 (unnamed assert)
ok 3 (unnamed assert)
ok 4 (unnamed assert)
/Users/bent.cardan/reqshark/node-nanomsg/node_modules/tape/index.js:75
        throw err
              ^
Error: stream.push() after EOF
    at readableAddChunk (_stream_readable.js:149:15)
    at Socket.Readable.push (_stream_readable.js:126:10)
    at Socket.<anonymous> (/Users/bent.cardan/reqshark/node-nanomsg/lib/index.js:207:24)

we need to rewrite node.js survey sockets in light of updates to respondent/survey behavior in libnanomsg. note that what we're testing here calls

socket.survey(msg)

on a surveyor socket, which is different than calling

socket.send(msg)

this is an inconsistent area of our API that has little or nothing to do with the upgrade.

one problem i see here is that we're removing a listener that doesn't exist. I think this type specific API stuff was done back when libnanomsg only partially implemented the surveyor/respondent pattern.

since then the pattern's been improved by the clib, certainly in beta6 and possibly earlier in beta5. let's open a separate issue to track this and not let it prevent the upgrade

[reqshark]

hmm, wait why does the survey pass on node 0.12 before the upgrade...

15:35 $ node test/survey.js 
TAP version 13
# inproc socket survey
ok 1 (unnamed assert)
ok 2 (unnamed assert)
ok 3 (unnamed assert)
ok 4 (unnamed assert)

1..4
# tests 4
# pass  4

# ok

that means it has something to do with the upgrade and we'll need to revise our survey socket code

[reqshark]

returning null here inside the survey's recv operation ends of the stream properly instead of getting EOF

[reqshark]

@nickdesaulniers i went through our survey code and made a small change to how survey's recv returns.

now I think the problem had more to do with streams than survey actually, though it's still interesting to me that we only saw this behavior with the upgrade.

anyway, by pushing a value of undefined onto a stream we get unexpected behavior.

by pushing null we end the stream, which is what i think we intend to do here: https://github.com/nickdesaulniers/node-nanomsg/blob/beta6/lib/index.js#L187

[reqshark]

dude when you get back from the BrazilJS conf Porto Alegre, this PR needs some ❤️ ... after this is merged we need to fix iojs3.. nan 2... i was holding off on nan2 since this changes the clib slightly.

btw hope ur having a good time down there.. 🌴 this comment doesnt have anything to do with the PR but i was checking out your your conference slides and was wondering if you could pls explain:

Prefer task level parallelism (postMessage) to data level parallelism (SharedArrayBuffer)

[nickdesaulniers]

k let's do 🍻 brother!

[nickdesaulniers]

@reqshark , can you publish to npm? 🍤

[reqshark]

done. now npm install has beta6 for everything < iojs-v2.5.0.

We'll need to get nan2 updates done for v3+.